OpenAI Foundation Model Lab (2015)
infrastructure pace layer · 2015–ongoing
lifespan: 50 yrs · motor: push
Class card for the OpenAI nonprofit→capped-profit LP→PBC foundation-model lab, founded December 2015 by Altman, Musk, Brockman, Sutskever, Karpathy, Hoffman, and others with a $1B initial pledge. Mission: "ensure that AGI benefits all of humanity." Operational telos: LLM API dominance + ChatGPT user acquisition (DM-Day). OpenAI is the singularly load-bearing DM instance of the foundation-model-as-infrastructure typology — the machine that reframed AI from a research subdiscipline into a civilizational-scale project. Motor is push (DM default): OpenAI independently pushes capability breakthroughs (GPT-1 2018, GPT-2 2019, GPT-3 2020, GPT-4 2023, ChatGPT Nov 2022) without responding to a telos of Progress. Substrate is incorporeal + cognitive + semiotic: the load-bearing operating substrate is ML training pipelines, RLHF data, model weights, and the AGI narrative that structures industry attention. The heavy corporeal substrate (GPU clusters, datacenters) is externalized to machine:microsoft-azure-cloud (black_hole_dependency at compute layer). SCHEMA STRESS flag (atlas DM-36 §w0): nonprofit→capped-profit LP (March 2019) → PBC subsidiary (September 2024) legal mutation; encoded in organizational_history field (OrganizationalForm sub-model per Wave-6 atlas §3.5 finding 21 / SYNTHESIS §2.3). The nonprofit-vs-PBC tension constitutes a dm_current = late_modernity (energetic zombie) pathology: massive compute consumption + narrative coherence eroding post-Altman-firing 2023 + nonprofit-mission-vs-PBC-tension. heavy_or_light: light — the platform layer is light; heavy substrate (Azure GPU clusters) is held by Microsoft. has_interiority: true — deepest internal model-eval + RLHF telemetry of any DM AI machine. V0.2 GAP: tep6_ai_biotech is in TepParadigm enum as tep6_ai_biotech; atlas seed notes tep6[EXTRAP] for the canonical tep6 candidate — encoded as tep_paradigm_secondary with tep_paradigm_contested=true.
Machine type
incorporeal
Plasticity
plastic
Substrate
Wave source
wave-9-atlas
Inputs
- azure_compute_microsoft_investment
- gpu_cluster_compute_via_azure
- phd_researcher_labor
- rlhf_training_data_and_gpt_pretraining_corpus
Outputs
- gpt_model_series_and_api_ecosystem
- chatgpt_consumer_product_200m_users
- agi_narrative_structuring_ai_industry
- api_revenue_usd_per_year
Landscape pressures
- regulatory_fragmentation_agi_governance (70% intensity)
- compute_arms_race_capital_intensity (88% intensity)
- nonprofit_pbc_mission_tension (72% intensity)
Intra-era couplings
- depends_on machine:microsoft-azure-cloud · 0.95 CANON
- competitor_of machine:google-deepmind-ai-lab · 0.88 CANON
- competitor_of machine:chinese-ai-labs-collective · 0.72 CANON
- regulatory_target_of EU GDPR Regulatory Apparatus (2018–ongoing) · 0.68 CANON
- parasitic_extraction Wikipedia (2001) · 0.75 CANON
Cross-era couplings
- parasitic_extraction Post-Humboldtian Research University (1810) · 0.80
- substrate_provision National Electrical Grid (Insull / US Grid, 1882–ongoing) · 0.92
- substrate_provision Bell System / AT&T (1876–1984) · 0.72 CANON
- zombie_dependency Industrial-Era Patent System (1790) · 0.62 EXTRAP
State variables
Phase snapshots
Notable instances
- GPT-1 (Generative Pre-trained Transformer, Jun 2018) (2018) — 117M parameters; first OpenAI GPT-architecture release; demonstrated unsupervised pre-training + fine-tuning paradigm.
- GPT-2 (Feb 2019) (2019) — 1.5B parameters; initial 'too dangerous to release' framing (Feb 2019); full release Nov 2019; text generation quality i…
- GPT-3 (May 2020) (2020) — 175B parameters; first API-accessible frontier model; Codex (code) + Copilot (Jun 2021) derived from this base.
- DALL-E 2 (Apr 2022) (2022) — Diffusion-based image generation; demonstrated multimodal expansion beyond text; DALL-E 1 (Jan 2021) predecessor.
- ChatGPT (Nov 30 2022) (2022) — 100M users by Jan 2023 (fastest-growing consumer app in history); 200M weekly active users 2024; inflection in mainstrea…
- GPT-4 (Mar 2023) (2023) — Frontier multimodal model; bar exam top 10%; powers ChatGPT Plus + Enterprise; first model to show broad task generaliza…
- Sora (Feb 2024) (2024) — Video generation (text-to-video); demonstrates diffusion-transformer at video scale; Apple ChatGPT integration WWDC Jun …
Sources
- Stoller, Matt (2019). Goliath: The 100-Year War Between Monopoly Power and Democracy · 75%
- Heaven, Will Douglas (2023). Inside OpenAI (MIT Technology Review series) · 82%
- Bender, Emily M. and Gebru, Timnit et al. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? · 80%
- OpenAI (2019). OpenAI Charter (2018) and Partnership LP Announcement (2019) · 88%
- Atlas (Prime Radiant) (2026). research/09-atlas/dm-mm-industrial-stubs/findings.md (DM-36) · 85%
- Financial Times / The Information (2024). OpenAI revenue and Microsoft investment reporting · 72%